In this paper we study the problem of estimating the distance of a sound source from a single microphone recording in a room environment. The room effect cannot be separated from the problem without making assumptions about the properties of the source signal. Therefore, it is necessary to develop methods of distance estimation separately for different types of source signals. In this paper, we focus on speech signals. The proposed solution is to compute a number of statistical and source specific features from the speech signal and to use pattern recognition techniques to develop a robust distance estimator for speech signals. Experiments with a database of real speech recordings showed that the proposed model is capable of estimating source distance with acceptable performance for applications such as ambient telephony.
|Title of host publication||Proceedings of the 126th AES convention|
|Number of pages||12|
|Publication status||Published - 2009|
|Event||126th AES convention: make the right connections - München, Germany|
Duration: 7 May 2009 → 10 May 2009
|Conference||126th AES convention|
|Period||07/05/2009 → 10/05/2009|